LeXi AI, an Indian legal AI, topped AIBE among 5 leading AI models, ahead of GPT-5.5, Gemini 3.1 Pro, Claude Opus 4.8 & DeepSeek V3.2. - The Tribune

For much of the past decade, artificial‑intelligence progress has been measured by scale, but a new phase is emerging where the focus shifts to domain‑specific expertise. The All India Bar Examination (AIBE), a public benchmark used by the Bar Council of India to qualify aspiring advocates, provides a clear window into this shift. In the most recent evaluation, five AI systems were tested on the same AIBE 20 paper under identical conditions, with no specialized prompting.

LeXi AI, a platform built specifically for the Indian legal system, achieved the highest score, answering 94 of 95 questions correctly for an accuracy of 98.95%. It surpassed GPT‑5.5 (96.84%), Gemini 3.1 Pro (95.79%), Claude Opus 4.8 (90.53%), and DeepSeek V3.2 (89.47%). The benchmark is fully reproducible, as the paper and answer key are publicly available.

The result demonstrates that a model trained in depth on a single profession can outperform leading general‑purpose models on a transparent legal benchmark. This has implications for law firms, corporate legal teams, and courts that are increasingly exploring AI‑assisted research, drafting, and compliance. LeXi AI’s founder, Onkar Rana, emphasizes that lawyers need technology that understands Indian law’s language, structure, and responsibilities, rather than a general chatbot.

(Source：The Tribune)